CN-116559702-B - Battery health state prediction method and device, storage medium and electronic equipment

CN116559702BCN 116559702 BCN116559702 BCN 116559702BCN-116559702-B

Abstract

In the method and device for predicting the state of health of the battery, the storage medium and the electronic equipment provided by the invention, the characteristic extraction network in the battery state of health prediction model is used for extracting the associated characteristic data from the battery characterization information, the associated characteristic data is characteristic data of the association relation among various parameters of the deep mining battery, then the associated characteristic data is used as input data of the circulating neural network, and the circulating neural network processes the associated characteristic data and then outputs the state of health prediction data of the battery. In the process of predicting the state of health of the battery, the feature extraction network is used for extracting the associated feature data with deep association, and then the associated feature data is used as input data of the time cycle network, so that the accuracy of predicting the state of health of the battery is improved.

Inventors

ZHAO ZIHAO
ZHOU JIANJIE
LIU XINFENG
CAO XIAOHUI
QIAN CHAO
JIANG LULU

Assignees

阳光储能技术有限公司

Dates

Publication Date: 20260508
Application Date: 20230505

Claims (11)

1. A method for predicting a state of health of a battery, comprising: acquiring battery characterization information of a battery in a preset time range, wherein the battery characterization information comprises information of voltage, current, temperature and battery capacity; Extracting associated feature data from the battery characterization information by utilizing a feature extraction network in a pre-trained battery state of health prediction model; Processing the associated characteristic data by using a cyclic neural network in the battery health state prediction model to obtain health state prediction data of the battery; The training process of the battery health state prediction model comprises the following steps: acquiring battery experimental data; Inputting the battery experimental data into a state prediction model, and training the state prediction model until the state prediction model meets a preset convergence condition to obtain a pre-training model; Acquiring battery working condition data; and adjusting the network weight of the pre-training model by applying the battery working condition data to obtain a battery health state prediction model.
2. The method of claim 1, wherein extracting associated feature data from the battery characterization information using a feature extraction network in a pre-trained battery state of health prediction model comprises: invoking the feature extraction network to process the battery characterization information to obtain initial feature data; And inputting the initial characteristic data into a full-connection layer in the battery health state prediction model, so that the full-connection layer performs dimension reduction on the initial characteristic data to obtain associated characteristic data.
3. The method of claim 1, wherein the recurrent neural network is one of a long-short-term memory recurrent neural network, a gated recurrent unit, or a time convolutional neural network.
4. The method according to claim 1, wherein the processing the associated feature data using the recurrent neural network in the battery state of health prediction model to obtain the state of health prediction data of the battery comprises: and triggering the cyclic neural network to process the associated characteristic data based on preset network weights, and outputting the predicted data of the health state of the battery.
5. The method as recited in claim 1, further comprising: pruning is carried out on the battery health state prediction model, and a pruned battery health state prediction model is obtained.
6. The method of claim 1, wherein the inputting the battery experimental data into a state prediction model, training the state prediction model until the state prediction model meets a preset convergence condition, and obtaining a pre-training model comprises: Selecting target data from the battery experimental data; Inputting the target data into the state prediction model, so that the state prediction model processes the target data and outputs prediction information corresponding to the target data, wherein the prediction information comprises health prediction data and health degree change rate; Judging whether the state prediction model meets the convergence condition or not based on the prediction information and a preset supervision signal corresponding to the target data; when it is determined that the state prediction model does not satisfy the convergence condition, adjusting a network weight of the state prediction model based on the prediction information and the supervisory signal, and then returning to perform the step of selecting target data among the battery experimental data; and when the state prediction model is determined to meet the convergence condition, determining the state prediction model as a pre-training model.
7. The method of claim 6, wherein the causing the state prediction model to process the target data and output prediction information corresponding to the target data comprises: Invoking a feature extraction network in the state prediction model to process the target data and outputting extracted data; Performing dimension reduction processing on the extracted data by using a full connection layer in the state prediction model, and outputting dimension reduction data; Determining a health degree change rate based on the dimensionality reduction data, and processing the dimensionality reduction data by using a cyclic neural network in the state prediction model to output health prediction data; and determining the health degree change rate and the health prediction data as prediction information.
8. The method of claim 6, wherein the adjusting network weights of the state prediction model based on the prediction information and the supervisory signal comprises: based on the health degree change rate in the prediction information and the preset health degree change rate in the supervision signal, adjusting the network weight of the feature extraction network in the state prediction model; Based on the health prediction data in the prediction information and the preset health data in the supervision signal, adjusting the network weight of the cyclic neural network in the state prediction model and the network weight of the feature extraction network.
9. A battery state of health prediction apparatus, comprising: The first acquisition unit is used for acquiring battery characterization information of the battery in a preset time range, wherein the battery characterization information comprises information of voltage, current, temperature and battery capacity; The extraction unit is used for extracting associated feature data from the battery characterization information by utilizing a feature extraction network in the battery state of health prediction model which is trained in advance; the processing unit is used for processing the associated characteristic data by utilizing a cyclic neural network in the battery health state prediction model to obtain health state prediction data of the battery; the second acquisition unit is used for acquiring battery experimental data; The training unit is used for inputting the battery experimental data into a state prediction model, and training the state prediction model until the state prediction model meets a preset convergence condition to obtain a pre-training model; the third acquisition unit is used for acquiring battery working condition data; and the adjusting unit is used for applying the battery working condition data to adjust the network weight of the pre-training model to obtain a battery health state prediction model.
10. A storage medium comprising stored instructions, wherein the instructions, when executed, control a device in which the storage medium is located to perform the method of predicting a state of health of a battery as claimed in any one of claims 1-8.
11. An electronic device comprising a memory, and one or more instructions, wherein the one or more instructions are stored in the memory and configured to perform the method of predicting a battery state of health of any one of claims 1-8 by one or more processors.

Description

Battery health state prediction method and device, storage medium and electronic equipment Technical Field The present invention relates to the field of battery technologies, and in particular, to a method and apparatus for predicting a battery state of health, a storage medium, and an electronic device. Background With the popularization of new energy, batteries are used as energy storage systems of automobiles, small and medium-sized unmanned aerial vehicles and various small and medium-sized machine equipment. Performance degradation is inevitable as the battery life increases. To better assess the useful life of the energy storage system, a prediction of battery state of health (SOH, state OfHealth) is required. At present, a battery SOH is usually predicted by extracting physical discrete features, such as voltage difference, temperature change rate, voltage inflection point kurtosis and the like, from a charge-discharge curve of the battery, and then predicting the SOH through the physical discrete features by a machine learning method, wherein the internal relation of each variable of the battery cannot be associated in the traditional battery SOH prediction process, so that the accuracy of the predicted battery SOH is not high. Disclosure of Invention In view of the above, embodiments of the present invention provide a method and apparatus for predicting a battery state of health, a storage medium, and an electronic device, where the method and apparatus use feature extraction network deep mining to obtain associated feature data, and use the feature data as input data of a recurrent neural network, so that the recurrent neural network can predict the battery state of health according to the feature data, and effectively improve accuracy of prediction. In order to achieve the above object, the embodiment of the present invention provides the following technical solutions: A method of predicting a state of health of a battery, comprising: acquiring battery characterization information of a battery in a preset time range, wherein the battery characterization information comprises information of voltage, current, temperature and battery capacity; Extracting associated feature data from the battery characterization information by utilizing a feature extraction network in a pre-trained battery state of health prediction model; And processing the associated characteristic data by using a cyclic neural network in the battery health state prediction model to obtain the health state prediction data of the battery. In the above method, optionally, the extracting, by using a feature extraction network in a pre-trained battery state of health prediction model, associated feature data from the battery characterization information includes: invoking the feature extraction network to process the battery characterization information to obtain initial feature data; And inputting the initial characteristic data into a full-connection layer in the battery health state prediction model, so that the full-connection layer performs dimension reduction on the initial characteristic data to obtain associated characteristic data. In the above method, optionally, the recurrent neural network is one of a long-short-term memory recurrent neural network, a gated recurrent unit and a time convolution neural network. In the above method, optionally, the processing the associated feature data by using a recurrent neural network in the battery health status prediction model to obtain health status prediction data of the battery includes: and triggering the cyclic neural network to process the associated characteristic data based on preset network weights, and outputting the predicted data of the health state of the battery. The method, optionally, the training process of the battery state of health prediction model includes: acquiring battery experimental data; Inputting the battery experimental data into a state prediction model, and training the state prediction model until the state prediction model meets a preset convergence condition to obtain a pre-training model; Acquiring battery working condition data; and adjusting the network weight of the pre-training model by applying the battery working condition data to obtain a battery health state prediction model. The method, optionally, further comprises: pruning is carried out on the battery health state prediction model, and a pruned battery health state prediction model is obtained. In the above method, optionally, the inputting the battery experimental data into a state prediction model, training the state prediction model, until the state prediction model meets a preset convergence condition, obtaining a pre-training model includes: Selecting target data from the battery experimental data; Inputting the target data into the state prediction model, so that the state prediction model processes the target data and outputs prediction information corresponding to the target data,